zkFMI
日本語

Roadmap

Ordered from the status documents, the use-case survey and the open questions in the position papers. There are no dates, because none has been promised, and the items are grouped by what they depend on rather than by when they might happen.

Update: 2026-09-12. Current non-PQC notes use Triptych-based membership and additively homomorphic commitments. Native confidential-asset execution and fresh-network hash/code-based PQC proof research have advanced; operational PQC notes and the full venue path remain incomplete. Historical measurements below retain their original constructions and environments. Note execution evidence · Current PQC work

Done and accepted

  • Protocol core in Rust, measured

    zkPI v2 with threshold range proofs and fixed vectors; DeFMI accounts, notes, tags, netting, limits, waterfall, PvP, reconciliation, scoped viewing, vetting; Bulletproof backend; DeKYX, DeCCP, Aethel crate workspaces.

  • Native Avalanche L1 with five validators, one host

    Atomic multi-RFQ settlement, replay rejection, restart recovery, state-root agreement. The QOMM path settles from seven MP-SPDZ processes with no additional signature.

  • OCLOB distributed path to the L1, one host

    Participant-side splitting, encrypted direct delivery to seven containers, 5-of-7 arrival certificates, MPC matching, maker and taker reservations, DvP with restart and replay checks.

  • Binding and accountability mechanisms

    Matched field at 2.00× traffic; sound per-party input check; share-locating decoder in the engine; robust reconstruction at n=9 measured on two machines.

  • Two review rounds on the record

    Soundness findings fixed and documented; retracted claims listed; misses kept.

  • Land the post-quantum integration on main

    Hybrid Ed25519 + ML-DSA-65 signatures and X25519 + ML-KEM-768 key exchange across all eight repositories passed a single-host native acceptance on 2026-09-06 and again on 2026-09-07 with the zero-relation opening fix. The DeFMI ⇄ QOMM dependency cycle was broken the same day by moving the stack crates to zkPI, every repository is pinned by immutable revision, and all eight are on main. Overhead was measured on 2026-09-08 on the two deployed OCLOB images; after the resident runner and the once-per-process receipt check, the hybrid image is 177 ms and 202 ms faster per order than the classical one. See Post-quantum migration.

Next, by dependency

  • Threshold settlement authority for OCLOB

    Replace the single research gateway that opens the participant's fixed-length settlement authority with threshold decryption or joint zkPI generation by the MPC nodes. The P0 item that keeps OCLOB off real assets.

  • Connect confidential-asset notes to venues

    The new native path has executed issuance, reservation, partial fills, refunds, claim redemption and recovery after restart. Connect existing QOMM/OCLOB callers to the complete confidential wrapper, then verify actual application inputs and independent approval and custody.

  • Atomic admit-pledge-limit-pay

    Grant and admit are separate calls and no state locks pledged collateral. Written requirement, not built.

  • Persistent state generations

    Encrypted queues, rate-limiter counters and nullifier ledgers in authenticated storage across restart, so a rollback cannot re-enable a replay and a venue run as two processes does not double every allowance.

  • Production surface

    mTLS, per-tenant authorisation, OpenAPI, rate limits, audit logs, external KMS/HSM, key rotation, backup and recovery. Currently demo APIs only.

  • Offline phase among the nodes

    The online phase is measured at 16% of the bytes with a trusted dealer; a real preprocessing protocol among seven nodes is not built.

  • Bias correction for differential-privacy disclosure

    DP disclosure measured worse than none from upward bias in a non-linear statistic; the in-MPC sampler needs an exact unbounded construction before a pure-DP claim.

  • Reference-price correction for slow committees

    The winner does not depend on the reference and the price is affine in it; the implementation still reveals a price as old as the computation.

Products, in the order the use-case survey recommends

  • 1. Fund subscription and redemption

    Separate keys for issuer, administrator, investor and KYC; eligibility, cap, reservation and issuable units bound to one zkPI; cash and unit notes updated together; over-subscription, expiry, replay and one-leg tests.

  • 2. Collateral and margin

    Eligibility, haircut and concentration as approved circuits; coverage proved over a secret portfolio; substitution and return; concurrent-call ordering under per-entity caps. The strongest research topic.

  • 3. Two-currency PvP

    Two issuers and two reservations in one instruction; outage, expiry, restart and resend tests; ISO 20022 in an external adapter.

  • 4. External ledger connection

    Mock fund, collateral and bank ledgers as independent processes; signed reservations and receipts; an explicit list of which faults break atomicity.

Blocked on things outside the code

  • Independent operators on a WAN

    Seven MPC operators, five or more validators, separate administrators, key stores and failure domains. The largest remaining unknown, and the one the risk register says is hardest.

  • Operational private note relations and security

    Bounded four-account collaborative proofs and canonical state updates have research execution evidence. The remaining work binds private note membership, spending authority, output encryption, value and asset conservation, and the complete venue lifecycle into one valid relation, with external security validation of the exact construction.

  • Legal finality and register connection

    A book-entry register, a CSD, a central-bank or designated cash leg, and a supervisor's disclosure workflow. In the EU and UK the natural home is non-equity RFQ; in Switzerland the ledger entry can be the right.

  • Independent audit

    Cryptographic, Rust state-machine, host-VM and integration audit by a third party. Two internal review rounds are not that.

Open research questions

  1. A publicly verifiable, linearly homomorphic commitment over F_p that opens many times on a random oracle only.
  2. Comparison without slack in a field that also holds a group order, for seven-party Shamir.
  3. The honest-majority price of public accountability.
  4. A formal security definition and proof for OCLOB, including communication leakage and selective halting.
  5. Whether Prime Match's two-round malicious comparison transfers to seven-party Shamir.